If you already had a recording of two separate sound sources mixed, where one source was a person singing and the second source was a piano playing. Could ICA be used to separate the recording into 2 distinct sound sources? If so, how would ICA be used to do this?
Yes, this can be done: here’s a paper from 2015 indicating recent developments and there’s a handy dandy tutorial here to get you started using scikit-learn’s ICA implementation. Enjoy!